Classifying post-traumatic stress disorder using the magnetoencephalographic connectome and machine learning.
Jing Zhang1,2, J Don Richardson3,4, Benjamin T Dunkley5,6,7
1Department of Diagnostic Imaging, Hospital for Sick Children, Toronto, ON, Canada. jing.zhang@sickkids.ca.
Scientific Reports
|April 5, 2020
Summary
Magnetoencephalography (MEG) neural synchrony, analyzed with machine learning, can objectively identify post-traumatic stress disorder (PTSD). This study developed a robust computational framework for PTSD classification using brain connectivity patterns.
Area of Science:
- Neuroscience
- Computational Psychiatry
- Biomarker Discovery
Background:
- Conventional post-traumatic stress disorder (PTSD) diagnosis relies on subjective methods.
- Objective biomarkers for PTSD are crucial for clinical and research applications.
- Magnetoencephalography (MEG) measures macroscopic neural circuits and has shown potential for PTSD assessment.
Purpose of the Study:
- To develop and validate a machine learning framework using MEG neural synchrony for objective PTSD classification.
- To identify specific neural synchrony patterns indicative of combat-related PTSD.
- To establish a computational model for classifying mental health conditions using brain connectome data.
Main Methods:
- Employed a machine learning classification framework integrating Support Vector Machine (SVM) with recursive random forest feature selection (CV-SVM-rRF-FS) on MEG data.
- Analyzed neural synchrony across five frequency bands to identify key features for distinguishing PTSD from controls.
- Utilized independent partial least squares discriminant analysis to assess model bias.
Main Results:
- The CV-SVM-rRF-FS identified minimal, frequency-specific neural synchrony 'edges' serving as PTSD signatures.
- Selected features align with known PTSD pathophysiology.
- The final SVM models achieved high classification performance (Area Under Curve up to 0.9), demonstrating robustness against a trauma-exposed control group.
Conclusions:
- Machine learning analysis of MEG neural synchrony offers a robust and objective method for PTSD classification.
- The developed computational framework shows promise for classifying other mental health disorders using MEG connectome profiles.
- This approach advances the search for objective biomarkers in psychiatry.
